% Script for determining colony doubling time from plate reader growth data.
%
% Users of the script should update the following variables:
%     FILENAME: A tab-delimited text file with only the data and a single header
%         row with well names.
%     INTERVAL: Minutes between readings.
%     FLATTEN_FIRST_N_MINUTES: The number of readings to ignore at the
%         beginning of the script.
%
% The output is written to a new file in the same location as the input
% a text file, with extension '.analyzed_growth.txt'. This can be imported into
% Excel as a tab-delimited file.
%
% Written by Jaron Mercer based on original implementation by Harris Wang.
% Updated June 2013 by Gleb Kuznetsov.

%AJR NOTES:
%Code looks for the first max slope.  When it finds it, it then starts counting a number of intervals (set below - see note below) until it finds a second max slope.  If the second max slope is within this time window, it will take that into account and choose the bigger of the two.  If it doesn't fall within this time window, it will  not find that second hump or second max slope.  However, you can change this window (see my note below).  
%For all DTs calculated in paper, intervals_since_greatest was set to > 24.  To look at the whole curve, I set it to 240 (for F and W suppressor containing strains).

clear;
close all;


%%% Set inputs.

% Windows style filename placeholder.
% FILENAME = 'C:\Users\Marc\Documents\0Harvard\Church Rotation\Church Laboratory Notebook\0rEcoli\Plate Reader Data\2013.08.20.Agg.Ess.Rib.double.mut.growth.K+S.txt';

% Unix style filename placeholder.
FILENAME = '20160824_GFPgrowth_03_matlab.txt';

% Minutes separating each reading.
INTERVAL = 10;

% Sometimes the growth data shows irregular behavior at the beginning.
% We'll copy the value at the following value to all previous values
% to avoid getting an erroneous reading.
% Set this to 0 if you don't want any flattening.
FLATTEN_FIRST_N_MINUTES = 0;


%%% Begin processing

input = importdata(FILENAME, '\t', 1);

% Matrix where rows are consecutive time measurements and each column
% corresponds to a well.
data = input.data;

% Well names.
headers = input.colheaders;

% The number of wells.
num_wells = size(data, 2);

% Flatten the data. See comment for FLATTEN_FIRST_N_MINUTES above.
if FLATTEN_FIRST_N_MINUTES > 0
  flatten_first_n_observations = FLATTEN_FIRST_N_MINUTES / INTERVAL;
  for well = 1:num_wells
    % Copy the value at the nth position to all previous positions.
    data(1:flatten_first_n_observations, well) = ...
        data(flatten_first_n_observations, well);
  end
end


%% Plot data.
whos data;
plot(data);
xlabel('Time (min/10)');
ylabel('OD 600');
title(strcat(headers{1,1}, ' - ', headers{1,size(headers,2)}));


%% Plot log of data.
ln_data = log (data);
ln_data(isinf(ln_data)) = NaN;
ln_data = real(ln_data);
figure (2); plot(ln_data);
xlabel('Time (min/10)');
ylabel('ln(OD 600)');
title(strcat('ln(', headers{1,1}, ' - ', headers{1,size(headers,2)}, ')'));


%%% Find the linear portion of ln(OD 600). Then do linear regression.
%
% We dynamically search for the limits of the linear portion of the log-OD
% plot. We do this by performing a linear regression on a sliding interval
% window, and keeping track of the window with the greatest slope.
%
% This is the main "algorithmic" part of the script and the primary change
% Jaron Mercer made to Harris Wang's version of this script.

doubleTs = zeros(size(ln_data,2),1);
rSqrs = zeros(size(ln_data,2),1);
maxODs = zeros(size(ln_data,2),1);
deltas = zeros(size(ln_data,2),1);
starts = zeros(size(ln_data,2),1);
warnings = zeros(size(ln_data,2),1);

for well = 1:num_wells
    % Initialize state variables.
    maxSlope = 0;
    maxR = 0;
    maxDelta = 0;
    maxStart = 0;

    % Find the greatest slope.
    intervals_since_greatest = 0;
    for delta = 5:7
        for start = 1:(size(ln_data,1)-delta)
            % We expect the greatest interval early on so no need to go more
            % than 2 hours past max.
            if intervals_since_greatest > 24 %This is how many intervals past the first max it will search through before the code shuts down.
                break
            end

            x = (linspace(start, start + delta - 1, delta))';
            y = ln_data(start: start + delta - 1, well);
            line = polyfit(x,y,1); % returns 1x2 matrix: [slope, y-intercept]

            if line(1,1) > maxSlope
                maxSlope = line(1,1);
                maxR = corrcoef(x,y);
                maxDelta = delta;
                maxStart = start;
                intervals_since_greatest = 0;
            else
                intervals_since_greatest = intervals_since_greatest + 1;
            end
        end
    end

    % Save output data.
    doubleTs(well, 1) = (log(2) / maxSlope) * INTERVAL;
    rSqrs(well, 1) = (maxR(1, 2)) ^ 2; %save r-squared
    maxODs(well, 1) = max(data(:, well));
    starts(well, 1) = maxStart * INTERVAL;
    deltas(well, 1) = maxDelta;

    % Output a warning for low r-squared values.
    if rSqrs(well, 1) < .99
        disp(strcat('Warning: low confidence on well --', headers(1, well)));
        figure(3); hold  on; plot(ln_data(:, well));
        warnings(well, 1) = 1;
    end
end


%%% Save data to a tab delimited text file.

output_filename = strcat(FILENAME(1:size(FILENAME, 2) - 4), '.analyzed_growth.txt');

output_data = [headers' num2cell(doubleTs) num2cell(rSqrs) num2cell(maxODs) num2cell(starts) num2cell(deltas) num2cell(warnings)];

fid = fopen(output_filename, 'w');

% Write the header row.
fprintf(fid, 'id\twell\tdoubling_time\tr_sqrd\tmax_OD\tstart_time_min\tdelta\twarnings\n');

% Iterate through the well data, writing one row at a time.
for well = 1:num_wells
  fprintf(fid, '%d\t%s\t%f\t%f\t%f\t%f\t%f\t%f\n', well, headers{well}, doubleTs(well), rSqrs(well), maxODs(well), starts(well), deltas(well), warnings(well));
end

fclose(fid);